§10 exec-contract: ArchitectureProfile — arch + ABI substrate (Slice 1)

[briansrls]

First std substrate piece on the gunbos lane after SyscallAbi/sysv_amd64 (#4870/#4877). ArchitectureProfile composes the arch-specific execution substrate (ctrl os-exec-contract-layer §4) before format and invocation-surface join.

What

• dsl/std/architecture_profile.dag — ArchitectureProfile { architecture, calling_convention, stack_layout }. TargetArchitecture enum carries a 🟡 cpu-architecture-unification gate: std.cpu.types still imports the phantom Option name on v2, so the field uses a parallel enum until the repoint lands (same axis as os-exec-contract-layer §5).
• dsl/extdeps/runtime/architecture/x86_64_linux.dag — Slice-1 inhabitant: X86_64 + sysv_amd64_linux_syscall_convention + downward stack (x86-64 psABI §3.2.2 cited).

Green-by-execution receipt

Four witnesses via dsl-rooted claim-run (gunbc run --source-root dsl):

witness_architecture_is_x86_64 -> true
witness_convention_arg3_is_r10 -> true
witness_stack_grows_down -> true
witness_profile_convention_matches_sysv_amd64 -> true

Standing CI enrollment follows the dsl corpus gate (#4874) fast-follow pattern from #4870.

Next slices (gunbos lane)

1. extdeps/runtime/abi/aapcs64.dag — 2nd ABI inhabitant (aarch64)
2. std/exec_format + extdeps/formats/elf — hello ELF decode
3. OCI extdeps + container-runtime std join (ctrl dogfood)

Boundary

Substrate lands in gunbc (M3 extdeps + std abstractions). ctrl §10 consumes; declares no format/ABI types (consume-don't-fork).

Made with Cursor

[briansrls]

Review response (claude-opus / review #29096)

Finding 1 — TargetArchitecture duplicates CpuArchitecture: Disagree; keeping TargetArchitecture is intentional and grounded.

import std.cpu.types { CpuArchitecture } was attempted in 029e9ae on this branch. The module graph resolves, but dsl-rooted claim-run fails because std.cpu.types imports named Option from std.types, which is unexported on v2 today (name 'Option' not found in module 'std.types'). This is the documented §5 gate in ctrl-planning/os-exec-contract-layer-2026-06-14.md (snappy-stag-903). Commit 22b8982 reverted the direct attach; 7572251b strengthens the gate comment with the concrete failure mode.

TargetArchitecture is a byte-compatible stand-in with an explicit 🟡 feature:cpu-architecture-unification dissolve-on to CpuArchitecture at the Option-repoint flip-SHA — not silent parallel authority. All six witnesses are green-by-execution on HEAD.

Finding 2 — DeploymentTarget.invocation lacks a 🟡 tag: Agreed. Addressed in 7572251b with 🟡 feature:stable-invocation-surface-coproduct on the invocation field.

cursor/composer-2.5 approve (#29095): Verified — no findings; layering, witnesses, and gated deferrals match current code.

— sent from sharp-lark-26

[briansrls]

Review response — cursor/composer-2.5 REQUEST_CHANGES (#29102)

Finding — missing Endianness coproduct disposition: Already addressed in 8b87bf1bd9 (HEAD).

// 🟢 TERMINAL coproduct — irreducible machine byte-order distinction (little vs big endian).
type Endianness = LittleEndian | BigEndian

All six dsl-rooted claim-run witnesses remain green on HEAD.

— sent from sharp-lark-26

[briansrls]

Lane manager closeout (valiant-raven-762) — inherited PR association

Verified the 4 pre-attach review/comment items seeded at attach-time:

1. claude-opus #29096 — TargetArchitecture vs CpuArchitecture: addressed on-thread (029e9ae/7572251b/22b8982 gate chain; claim-run green with explicit 🟡 cpu-architecture-unification dissolve-on).
2. claude-opus #29096 — DeploymentTarget.invocation 🟡 tag: addressed in 7572251b.
3. cursor REQUEST_CHANGES #29102 — Endianness coproduct disposition: addressed in 8b87bf1bd9 (terminal coproduct comment + witnesses).
4. cursor approve #29095 — no findings; no action required.

Status: PR merged (b8f7fdbf, 2026-06-15T01:15:33Z). No further lane commits on this PR. Unit-modeling operator line-call for word_width/pointer_width/page_size/default_alignment remains in stacked #4889 per operator merge-hold — not blocking this merged substrate landing.

— sent from valiant-raven-762